1
  2
  3
  4
  5
  6
  7
  8
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
// This file is part of Substrate.

// Copyright (C) 2020-2021 Parity Technologies (UK) Ltd.
// SPDX-License-Identifier: GPL-3.0-or-later WITH Classpath-exception-2.0

// This program is free software: you can redistribute it and/or modify
// it under the terms of the GNU General Public License as published by
// the Free Software Foundation, either version 3 of the License, or
// (at your option) any later version.

// This program is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU General Public License for more details.

// You should have received a copy of the GNU General Public License
// along with this program. If not, see <https://www.gnu.org/licenses/>.

//! Substrate service tasks management module.

use crate::{
	config::{JoinFuture, TaskExecutor, TaskType},
	Error,
};
use exit_future::Signal;
use futures::{
	future::{join_all, pending, select, try_join_all, BoxFuture, Either},
	sink::SinkExt,
	Future, FutureExt, StreamExt,
};
use log::{debug, error};
use prometheus_endpoint::{
	exponential_buckets, register, CounterVec, HistogramOpts, HistogramVec, Opts, PrometheusError,
	Registry, U64,
};
use sp_utils::mpsc::{tracing_unbounded, TracingUnboundedReceiver, TracingUnboundedSender};
use std::{panic, pin::Pin, result::Result};
use tracing_futures::Instrument;

mod prometheus_future;
#[cfg(test)]
mod tests;

/// An handle for spawning tasks in the service.
#[derive(Clone)]
pub struct SpawnTaskHandle {
	on_exit: exit_future::Exit,
	executor: TaskExecutor,
	metrics: Option<Metrics>,
	task_notifier: TracingUnboundedSender<JoinFuture>,
}

impl SpawnTaskHandle {
	/// Spawns the given task with the given name.
	///
	/// Note that the `name` is a `&'static str`. The reason for this choice is that statistics
	/// about this task are getting reported to the Prometheus endpoint (if enabled), and that
	/// therefore the set of possible task names must be bounded.
	///
	/// In other words, it would be a bad idea for someone to do for example
	/// `spawn(format!("{:?}", some_public_key))`.
	pub fn spawn(&self, name: &'static str, task: impl Future<Output = ()> + Send + 'static) {
		self.spawn_inner(name, task, TaskType::Async)
	}

	/// Spawns the blocking task with the given name. See also `spawn`.
	pub fn spawn_blocking(
		&self,
		name: &'static str,
		task: impl Future<Output = ()> + Send + 'static,
	) {
		self.spawn_inner(name, task, TaskType::Blocking)
	}

	/// Helper function that implements the spawning logic. See `spawn` and `spawn_blocking`.
	fn spawn_inner(
		&self,
		name: &'static str,
		task: impl Future<Output = ()> + Send + 'static,
		task_type: TaskType,
	) {
		if self.task_notifier.is_closed() {
			debug!("Attempt to spawn a new task has been prevented: {}", name);
			return
		}

		let on_exit = self.on_exit.clone();
		let metrics = self.metrics.clone();

		// Note that we increase the started counter here and not within the future. This way,
		// we could properly visualize on Prometheus situations where the spawning doesn't work.
		if let Some(metrics) = &self.metrics {
			metrics.tasks_spawned.with_label_values(&[name]).inc();
			// We do a dummy increase in order for the task to show up in metrics.
			metrics.tasks_ended.with_label_values(&[name, "finished"]).inc_by(0);
		}

		let future = async move {
			if let Some(metrics) = metrics {
				// Add some wrappers around `task`.
				let task = {
					let poll_duration = metrics.poll_duration.with_label_values(&[name]);
					let poll_start = metrics.poll_start.with_label_values(&[name]);
					let inner =
						prometheus_future::with_poll_durations(poll_duration, poll_start, task);
					// The logic of `AssertUnwindSafe` here is ok considering that we throw
					// away the `Future` after it has panicked.
					panic::AssertUnwindSafe(inner).catch_unwind()
				};
				futures::pin_mut!(task);

				match select(on_exit, task).await {
					Either::Right((Err(payload), _)) => {
						metrics.tasks_ended.with_label_values(&[name, "panic"]).inc();
						panic::resume_unwind(payload)
					},
					Either::Right((Ok(()), _)) => {
						metrics.tasks_ended.with_label_values(&[name, "finished"]).inc();
					},
					Either::Left(((), _)) => {
						// The `on_exit` has triggered.
						metrics.tasks_ended.with_label_values(&[name, "interrupted"]).inc();
					},
				}
			} else {
				futures::pin_mut!(task);
				let _ = select(on_exit, task).await;
			}
		};

		let join_handle = self.executor.spawn(future.in_current_span().boxed(), task_type);

		let mut task_notifier = self.task_notifier.clone();
		self.executor.spawn(
			Box::pin(async move {
				if let Err(err) = task_notifier.send(join_handle).await {
					error!("Could not send spawned task handle to queue: {}", err);
				}
			}),
			TaskType::Async,
		);
	}
}

impl sp_core::traits::SpawnNamed for SpawnTaskHandle {
	fn spawn_blocking(&self, name: &'static str, future: BoxFuture<'static, ()>) {
		self.spawn_blocking(name, future);
	}

	fn spawn(&self, name: &'static str, future: BoxFuture<'static, ()>) {
		self.spawn(name, future);
	}
}

/// A wrapper over `SpawnTaskHandle` that will notify a receiver whenever any
/// task spawned through it fails. The service should be on the receiver side
/// and will shut itself down whenever it receives any message, i.e. an
/// essential task has failed.
#[derive(Clone)]
pub struct SpawnEssentialTaskHandle {
	essential_failed_tx: TracingUnboundedSender<()>,
	inner: SpawnTaskHandle,
}

impl SpawnEssentialTaskHandle {
	/// Creates a new `SpawnEssentialTaskHandle`.
	pub fn new(
		essential_failed_tx: TracingUnboundedSender<()>,
		spawn_task_handle: SpawnTaskHandle,
	) -> SpawnEssentialTaskHandle {
		SpawnEssentialTaskHandle { essential_failed_tx, inner: spawn_task_handle }
	}

	/// Spawns the given task with the given name.
	///
	/// See also [`SpawnTaskHandle::spawn`].
	pub fn spawn(&self, name: &'static str, task: impl Future<Output = ()> + Send + 'static) {
		self.spawn_inner(name, task, TaskType::Async)
	}

	/// Spawns the blocking task with the given name.
	///
	/// See also [`SpawnTaskHandle::spawn_blocking`].
	pub fn spawn_blocking(
		&self,
		name: &'static str,
		task: impl Future<Output = ()> + Send + 'static,
	) {
		self.spawn_inner(name, task, TaskType::Blocking)
	}

	fn spawn_inner(
		&self,
		name: &'static str,
		task: impl Future<Output = ()> + Send + 'static,
		task_type: TaskType,
	) {
		let essential_failed = self.essential_failed_tx.clone();
		let essential_task = std::panic::AssertUnwindSafe(task).catch_unwind().map(move |_| {
			log::error!("Essential task `{}` failed. Shutting down service.", name);
			let _ = essential_failed.close_channel();
		});

		let _ = self.inner.spawn_inner(name, essential_task, task_type);
	}
}

impl sp_core::traits::SpawnEssentialNamed for SpawnEssentialTaskHandle {
	fn spawn_essential_blocking(&self, name: &'static str, future: BoxFuture<'static, ()>) {
		self.spawn_blocking(name, future);
	}

	fn spawn_essential(&self, name: &'static str, future: BoxFuture<'static, ()>) {
		self.spawn(name, future);
	}
}

/// Helper struct to manage background/async tasks in Service.
pub struct TaskManager {
	/// A future that resolves when the service has exited, this is useful to
	/// make sure any internally spawned futures stop when the service does.
	on_exit: exit_future::Exit,
	/// A signal that makes the exit future above resolve, fired on service drop.
	signal: Option<Signal>,
	/// How to spawn background tasks.
	executor: TaskExecutor,
	/// Prometheus metric where to report the polling times.
	metrics: Option<Metrics>,
	/// Send a signal when a spawned essential task has concluded. The next time
	/// the service future is polled it should complete with an error.
	essential_failed_tx: TracingUnboundedSender<()>,
	/// A receiver for spawned essential-tasks concluding.
	essential_failed_rx: TracingUnboundedReceiver<()>,
	/// Things to keep alive until the task manager is dropped.
	keep_alive: Box<dyn std::any::Any + Send + Sync>,
	/// A sender to a stream of background tasks. This is used for the completion future.
	task_notifier: TracingUnboundedSender<JoinFuture>,
	/// This future will complete when all the tasks are joined and the stream is closed.
	completion_future: JoinFuture,
	/// A list of other `TaskManager`'s to terminate and gracefully shutdown when the parent
	/// terminates and gracefully shutdown. Also ends the parent `future()` if a child's essential
	/// task fails.
	children: Vec<TaskManager>,
}

impl TaskManager {
	/// If a Prometheus registry is passed, it will be used to report statistics about the
	/// service tasks.
	pub fn new(
		executor: TaskExecutor,
		prometheus_registry: Option<&Registry>,
	) -> Result<Self, PrometheusError> {
		let (signal, on_exit) = exit_future::signal();

		// A side-channel for essential tasks to communicate shutdown.
		let (essential_failed_tx, essential_failed_rx) = tracing_unbounded("mpsc_essential_tasks");

		let metrics = prometheus_registry.map(Metrics::register).transpose()?;

		let (task_notifier, background_tasks) = tracing_unbounded("mpsc_background_tasks");
		// NOTE: for_each_concurrent will await on all the JoinHandle futures at the same time. It
		// is possible to limit this but it's actually better for the memory foot print to await
		// them all to not accumulate anything on that stream.
		let completion_future = executor
			.spawn(Box::pin(background_tasks.for_each_concurrent(None, |x| x)), TaskType::Async);

		Ok(Self {
			on_exit,
			signal: Some(signal),
			executor,
			metrics,
			essential_failed_tx,
			essential_failed_rx,
			keep_alive: Box::new(()),
			task_notifier,
			completion_future,
			children: Vec::new(),
		})
	}

	/// Get a handle for spawning tasks.
	pub fn spawn_handle(&self) -> SpawnTaskHandle {
		SpawnTaskHandle {
			on_exit: self.on_exit.clone(),
			executor: self.executor.clone(),
			metrics: self.metrics.clone(),
			task_notifier: self.task_notifier.clone(),
		}
	}

	/// Get a handle for spawning essential tasks.
	pub fn spawn_essential_handle(&self) -> SpawnEssentialTaskHandle {
		SpawnEssentialTaskHandle::new(self.essential_failed_tx.clone(), self.spawn_handle())
	}

	/// Send the signal for termination, prevent new tasks to be created, await for all the existing
	/// tasks to be finished and drop the object. You can consider this as an async drop.
	///
	/// It's always better to call and await this function before exiting the process as background
	/// tasks may be running in the background. If the process exit and the background tasks are not
	/// cancelled, this will lead to objects not getting dropped properly.
	///
	/// This is an issue in some cases as some of our dependencies do require that we drop all the
	/// objects properly otherwise it triggers a SIGABRT on exit.
	pub fn clean_shutdown(mut self) -> Pin<Box<dyn Future<Output = ()> + Send>> {
		self.terminate();
		let children_shutdowns = self.children.into_iter().map(|x| x.clean_shutdown());
		let keep_alive = self.keep_alive;
		let completion_future = self.completion_future;

		Box::pin(async move {
			join_all(children_shutdowns).await;
			completion_future.await;
			drop(keep_alive);
		})
	}

	/// Return a future that will end with success if the signal to terminate was sent
	/// (`self.terminate()`) or with an error if an essential task fails.
	///
	/// # Warning
	///
	/// This function will not wait until the end of the remaining task. You must call and await
	/// `clean_shutdown()` after this.
	pub fn future<'a>(
		&'a mut self,
	) -> Pin<Box<dyn Future<Output = Result<(), Error>> + Send + 'a>> {
		Box::pin(async move {
			let mut t1 = self.essential_failed_rx.next().fuse();
			let mut t2 = self.on_exit.clone().fuse();
			let mut t3 = try_join_all(
				self.children
					.iter_mut()
					.map(|x| x.future())
					// Never end this future if there is no error because if there is no children,
					// it must not stop
					.chain(std::iter::once(pending().boxed())),
			)
			.fuse();

			futures::select! {
				_ = t1 => Err(Error::Other("Essential task failed.".into())),
				_ = t2 => Ok(()),
				res = t3 => Err(res.map(|_| ()).expect_err("this future never ends; qed")),
			}
		})
	}

	/// Signal to terminate all the running tasks.
	pub fn terminate(&mut self) {
		if let Some(signal) = self.signal.take() {
			let _ = signal.fire();
			// NOTE: this will prevent new tasks to be spawned
			self.task_notifier.close_channel();
			for child in self.children.iter_mut() {
				child.terminate();
			}
		}
	}

	/// Set what the task manager should keep alive, can be called multiple times.
	pub fn keep_alive<T: 'static + Send + Sync>(&mut self, to_keep_alive: T) {
		// allows this fn to safely called multiple times.
		use std::mem;
		let old = mem::replace(&mut self.keep_alive, Box::new(()));
		self.keep_alive = Box::new((to_keep_alive, old));
	}

	/// Register another TaskManager to terminate and gracefully shutdown when the parent
	/// terminates and gracefully shutdown. Also ends the parent `future()` if a child's essential
	/// task fails. (But don't end the parent if a child's normal task fails.)
	pub fn add_child(&mut self, child: TaskManager) {
		self.children.push(child);
	}
}

#[derive(Clone)]
struct Metrics {
	// This list is ordered alphabetically
	poll_duration: HistogramVec,
	poll_start: CounterVec<U64>,
	tasks_spawned: CounterVec<U64>,
	tasks_ended: CounterVec<U64>,
}

impl Metrics {
	fn register(registry: &Registry) -> Result<Self, PrometheusError> {
		Ok(Self {
			poll_duration: register(HistogramVec::new(
				HistogramOpts {
					common_opts: Opts::new(
						"tasks_polling_duration",
						"Duration in seconds of each invocation of Future::poll"
					),
					buckets: exponential_buckets(0.001, 4.0, 9)
						.expect("function parameters are constant and always valid; qed"),
				},
				&["task_name"]
			)?, registry)?,
			poll_start: register(CounterVec::new(
				Opts::new(
					"tasks_polling_started_total",
					"Total number of times we started invoking Future::poll"
				),
				&["task_name"]
			)?, registry)?,
			tasks_spawned: register(CounterVec::new(
				Opts::new(
					"tasks_spawned_total",
					"Total number of tasks that have been spawned on the Service"
				),
				&["task_name"]
			)?, registry)?,
			tasks_ended: register(CounterVec::new(
				Opts::new(
					"tasks_ended_total",
					"Total number of tasks for which Future::poll has returned Ready(()) or panicked"
				),
				&["task_name", "reason"]
			)?, registry)?,
		})
	}
}